ci(package-c): build pytorch plugin without bundled runtime

[njzjz]

Summary

• build the C package in the manylinux_2_28 CUDA 12.9 image and install TensorFlow/PyTorch from dependency groups
• package the PyTorch backend plugin while excluding libtorch/CUDA runtime libraries from the tarball
• add package README/download_libtorch.sh guidance and update C package tests for external PyTorch runtime
• avoid CUDA 12.9 CCCL failures from -arch=all by using all-major as the CUDA 12.9 default

Tests

• ruff check .
• ruff format .
• git diff --check
• sh -n source/install/package_c.sh
• bash -n source/install/docker_package_c.sh
• bash -n source/install/docker_test_package_c.sh
• fork CI on bbcd58d: Build C library, Build C++, Build/upload to PyPI, CodeQL, Test C++, and Test Python all passed

Fork CI: https://github.com/njzjz/deepmd-kit/actions/runs/28547373402

Summary by CodeRabbit

• New Features
  • Expanded C library packaging and installation to support PyTorch in addition to TensorFlow and JAX.
  • When PyTorch is enabled, the package can generate a matching libtorch download helper and runtime setup guidance.
• Bug Fixes
  • Added stronger packaging verification to ensure required shared libraries are present and to prevent accidental bundling of PyTorch/CUDA runtime libraries.
• Documentation
  • Updated C library installation docs with PyTorch version/compatibility guidance and runtime environment setup (including LD_LIBRARY_PATH).

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 965ac2bd-7b78-4817-a4a2-e4b91c9b9175

📥 Commits

Reviewing files that changed from the base of the PR and between bbcd58d and 5a6ded6.

📒 Files selected for processing (2)

• source/install/package_c.sh
• source/lib/src/gpu/CMakeLists.txt

🚧 Files skipped from review as they are similar to previous changes (1)

• source/install/package_c.sh

---

📝 Walkthrough

Walkthrough

This PR updates the C library packaging pipeline for PyTorch runtime handling, adds container and CI checks around the packaged tarball, revises the installation guide, and changes the CUDA architecture default selection.

Changes

PyTorch-aware packaging pipeline

Layer / File(s)	Summary
Runtime dependency filtering source/api_c/CMakeLists.txt	Adds cached pre- and post-exclude regex variables, rebuilds the runtime library list, passes the computed values into the install script, and installs resolved runtime dependencies and packaged targets into the C library output tree.
Package script and libtorch helper generation source/install/package_c.sh	Updates path handling, probes Python and CUDA runtime inputs, parameterizes the CMake invocation, derives a PyTorch download URL when needed, and generates the README and download_libtorch.sh helper inside the staged C library archive.
Docker packaging container source/install/docker_package_c.sh	Switches to a configurable manylinux CUDA image, passes PyTorch and TensorFlow dependency settings into the container, and installs dependencies with uv before invoking package_c.sh.
Docker test runtime checks source/install/docker_test_package_c.sh	Uses a manylinux 2.28 image, builds and runs the C examples, and optionally validates bundled PyTorch runtime loading with ldd and ctypes.CDLL when CHECK_PYTORCH_RUNTIME is set.
CI validation and installation docs .github/workflows/package_c.yml, doc/install/install-from-c-library.md	Adds PyTorch runtime exclusion regexes, enables PyTorch packaging, verifies tarball contents, records a download URL note, turns on runtime checks, and updates the install guide for PyTorch backend support and libtorch helper usage.

CUDA architecture default

Layer / File(s)	Summary
CUDA toolkit version-based architecture default source/lib/src/gpu/CMakeLists.txt	Moves CUDA toolkit discovery before architecture defaulting and sets the default architecture to all-major for CUDA toolkit versions 12.9 through below 14.0, otherwise all.

Estimated code review effort: 4 (Complex) | ~60 minutes

Possibly related PRs

• deepmodeling/deepmd-kit#5475 — Also changes source/api_c/CMakeLists.txt in the C packaging path.

Suggested reviewers: wanghan-iapcm, iProzd

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately captures the main change: packaging the PyTorch C plugin without bundling its runtime libraries.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 3

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@source/install/package_c.sh`:
- Around line 119-215: The PyTorch runtime URL derivation in package_c.sh can
fail silently when ENABLE_PYTORCH is true but PYTHON_BIN is missing or the torch
version heuristic does not produce a supported URL. Add a clear warning or
diagnostic in the URL-generation block before the README and
download_libtorch.sh generation is skipped, using the existing
PYTORCH_RUNTIME_DOWNLOAD_URL, ENABLE_PYTORCH, and PYTHON_BIN checks to explain
why no PyTorch runtime package was emitted. Keep the warning tied to the same
install-script flow so local/manual runs surface the missing libtorch packaging
immediately.
- Around line 122-138: The libtorch URL generation in the torch version handling
logic is stripping nightly `.devYYYYMMDD` data by splitting on `.dev`, so fix
the version parsing in the package_c script to preserve the full nightly version
string when building the download URL. Update the `version`/`variant` extraction
path around the `torch.__version__` and `torch.version.cuda` handling so nightly
builds keep their `.dev` suffix while still producing the correct `cpu` or `cu*`
libtorch archive name.

In `@source/lib/src/gpu/CMakeLists.txt`:
- Around line 9-17: The CUDA architecture fallback in CMakeLists.txt is too
narrow because the `all-major` branch in the `find_package(CUDAToolkit)` /
`CMAKE_CUDA_ARCHITECTURES` logic stops at CUDA 12.x. Update the version check so
the `all-major` fallback also applies to CUDA 13.x, keeping the existing
`CUDAToolkit_VERSION` guard and the `set(CMAKE_CUDA_ARCHITECTURES ...)` logic in
place while extending the upper bound appropriately.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 3872c764-c654-40a6-9a12-928f465e34ab

📥 Commits

Reviewing files that changed from the base of the PR and between 42de67e and bbcd58d.

📒 Files selected for processing (7)

• .github/workflows/package_c.yml
• doc/install/install-from-c-library.md
• source/api_c/CMakeLists.txt
• source/install/docker_package_c.sh
• source/install/docker_test_package_c.sh
• source/install/package_c.sh
• source/lib/src/gpu/CMakeLists.txt

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.15%. Comparing base (ee30956) to head (5a6ded6).
⚠️ Report is 20 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #5706      +/-   ##
==========================================
- Coverage   81.20%   81.15%   -0.05%     
==========================================
  Files         978      990      +12     
  Lines      108858   111018    +2160     
  Branches     4139     4237      +98     
==========================================
+ Hits        88393    90101    +1708     
- Misses      18946    19390     +444     
- Partials     1519     1527       +8     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.